Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
bwaslo;234946 Wrote: I have done some interesting tests comparing 24bit/192kHz recorded music with a 16bit/44.1kHz version recorded from the same mix. I sample rate converted with the r8brain software so both were at 192kHz, then diffed them. Even with the (I assume) imperfect sample rate conversion used, it was rather shocking how very low the diff recording was. I'd like to repeat with other recordings (192kHz WAV files, with 44.1kHz equivalents aren't easy to come by, does anyone have a source?), but for now I am less enthused by the idea of high rate sampling than I had been. Seems reasonable. If 192/24 (or even 96/24) had been the leap equivalent to HDTV vs SDTV for example, we'd have made the switch by now... it just didn't turn out to be a compelling enough improvement. Maybe that Nyquist chap was onto something after all, and maybe those ringing filters aren't as bad as theory implies... ? -- Phil Leigh You want to see the signal path BEFORE it gets onto a CD/vinyl...it ain't what you'd call minimal... ...SB3+TACT+Altmann+MF DACXV3/Linn tri-amped Aktiv 5.1 system and some very expensive cables ;o) Phil Leigh's Profile: http://forums.slimdevices.com/member.php?userid=85 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
Thanks, but that is the one I did the one test with! -- bwaslo bwaslo's Profile: http://forums.slimdevices.com/member.php?userid=13548 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
Hi, I just ran into this thread, apparently several months late. I have to agree about some of the problems that some have had with my DiffMaker program. In some setups, a good clean null is achieved when it should be (i.e., when two recordings recorded the same way are diffed). In others, the null will be poor, except at the midpoint of the overlapped area of the two recordings. I recently ran into a good example of this myself. It seems the culprit is, as some here have suggested, varying clock speeds (I had thought it was Gain Drift, even modified the program to deal with that, but that wasn't it). Very small differences in the tracking of clock speeds between the source and the recorder will hurt the null. In some cases, the equipment used will track well, at least over enough time for the two recordings to be made. But too often they won't. For some tests (cables, capacitors, etc) this is easily dealt with, just use the same soundcard to emit the test sound track as is used to record with. That's the situation I mostly have worked with, which is why I didn't pick up on the importance of clock tracking (and also why I didn't highlight it in big red letters -- I'm just experimenting with this stuff myself, too, not dropping gold nuggets of wisdom here. And hey, I don't charge for this software!). I have made good diff nulls without clock tracking, but that appears to be a matter of luck. Much less than 1ppm tracking will be needed to get deep nulls, it turns out. I updated the help file and software notes to emphasize the issue of sample rate drift and matching. I have had extremely good results when I locked the soundcard (an ESI Juli@, used as an analog recorder)clock to a CD recorder's digital output. That is a good scenario, if it can be managed (it is easy with that sound card). I am working on a very-fine sample rate error detection and correction algorithm. It isn't difficult to do in general, the theory as most of you know is pretty straightforward, but getting it to operate within a reasonable amount of time is a problem. Maybe in DiffMaker v2 I'll have a way to do it faster. I have done some interesting tests comparing 24bit/192kHz recorded music with a 16bit/44.1kHz version recorded from the same mix. I sample rate converted with the r8brain software so both were at 192kHz, then diffed them. Even with the (I assume) imperfect sample rate conversion used, it was rather shocking how very low the diff recording was. I'd like to repeat with other recordings (192kHz WAV files, with 44.1kHz equivalents aren't easy to come by, does anyone have a source?), but for now I am less enthused by the idea of high rate sampling than I had been. -- bwaslo bwaslo's Profile: http://forums.slimdevices.com/member.php?userid=13548 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
bwaslo, Try these: http://www.soundkeeperrecordings.com/format.htm -- CPC CPC's Profile: http://forums.slimdevices.com/member.php?userid=12336 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I tried out the auido diffmaker, and it seems to be a pretty good program. If you check the help in one of the final chapters it makes mention that it best works if the DAC and ADC have the same clock. It does not account for dynamic variations of the sample frequency. Provided you look data over shorter periods and at the lower frequency ranges, this is not too much of a limitation. So the null that was found is likely the point where the waveforms were time correlated. If you estimate a difference of about 0.001 ppm between two recordings from the same source, that would be extremely good. I would not be surprised to see a variation of 1ppm of the same oscillator over time: even more just during warm up. I tried audio diffmaker to compare the output of an ipod to a toshiba laptop using the supplied pink noise burst. The plot is of the resulting Audio DiffMaker output. +---+ |Filename: ipod-noise.jpg | |Download: http://forums.slimdevices.com/attachment.php?attachmentid=3126| +---+ -- kgoulet kgoulet's Profile: http://forums.slimdevices.com/member.php?userid=12688 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I made a CD with the pink noise cited that came with the prog, had to pad it out 1s either side and convert to 44k1 using Audition to get it on the CD. The active part is very short. I then ran this through an Iriver portable CD player and recorded the signal using the Sound Devices 702. I've attached part of the residual. This time no minimum is detectable, though the residual is much lower. The pink noise is not steady noise - a change in tone colour is audible at several points in it and these changes correspond to the spikes, and are curiously repeatable across the channels - the CD source L and R channels are simly duplicated, but the replay and recording are two separate channels. My recording peaks at -33dBFS and the residual is say -76dBFS if you ignore the spikes, -51dBFS if you don't ignore them. System noise is slightly lower than the residual - you can just see the run-in where the noise was padded with silence before. Being charitable to audiodiffmaker the difference between the same thing repeated is -43dB relative to the original - if the spikes are included the difference is -25dB. I don't think this program is repeatable enough to show what it is supposed to show at all. I don't share the belief that all digital paths sound the same but I *do* think the same digital path will sound the same 10 seconds later. I have some scores in the cable and vibration-sensitive department I was hoping to settle with this (and I'm open to testing digital paths that sound apparently different :) However, as it stands this program isn't cutting it. If it *needs* wordclock synchronised across replay and recording systems then audiodiffmaker really ought to say so in big red letters. Though I can synch the SD to an external source and or send wordclock out of it I don't have have a CD player or squeezebox that can do that so I've reached the end of the road with it. As an independent observer I have *not* been able to replicate the results that program is claimed to show. +---+ |Filename: residual.jpg | |Download: http://forums.slimdevices.com/attachment.php?attachmentid=3128| +---+ -- ermine ermine's Profile: http://forums.slimdevices.com/member.php?userid=12613 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
opaqueice;219369 Wrote: In none of those cases did I get anything like what you're describing. Ahh, but did you do it blind? -- Patrick Dixon www.at-tunes.co.uk Patrick Dixon's Profile: http://forums.slimdevices.com/member.php?userid=90 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I tried a few years ago to write an application similar to audio diffmaker. It's open source http://wavm.sf.net There is even an example of a redigitized signal compared to the source .wav file. The application is not complete, but it shows that: - there's noise but it could be from the CD player (an old Sony car discman) or from the on-board A/D converter. - there is a gradual drift over time because I did not have a dynamic realignment where I readjust the alignment over time to account for CD player system clock error and A/D converter clock error. So there is the key problem: how to do the alignment: a) plain cross correlate only in time domain? This may weight the alignment more towards the larger amplitude low frequencies. b) Pick a few frequncies and monitor the phase between them? c) how to warp the time domains to minimize the error. d) how to cancel out common mode noise (common between the two sources) like 60/50Hz hum. This part can make the error appear to be larger than it is. If you use a single frequency, you may get more clues. ( http://wavb.sf.net is a utility for signle tone .wav files ). And //wava.sf.net does word by word compare between wav files. Audio diff does not appear to be open source, but working with single frequency tones maybe someone can gather some clues. -- kgoulet kgoulet's Profile: http://forums.slimdevices.com/member.php?userid=12688 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
It must have been an interesting challenge trying to write that. It sounds like fun. Could it be such a gradual drift that explains ermine's results? He used the same source (an SB1) each time, and still found a non-zero difference file, but with a partial null partway in. So if the clock rate of the SB1 was slightly different on the two recording sessions, that might account for his results. Seems odd, but maybe it's possible. And there's also the ADC clock - maybe that drifts for some reason? -- opaqueice opaqueice's Profile: http://forums.slimdevices.com/member.php?userid=4234 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
opaqueice;219539 Wrote: And there's also the ADC clock - maybe that drifts for some reason? The ADC clock will drift relative to the playout clock, even assuming the playout is jitterless. Replay clock is presumably determined by the SB1 or the server PC. However, I have used three independent ADC clocks made by widely differing manufacturers - Sound Devices, Toshiba (for the laptop) and Sony for the HiMD. They can't *all* be bad. It is, of course, possible that the SB1 DAC clock is a ropey PLL and it drifts and audiodiffmaker is correctly showing that temporal variation, in a similar vein to Omega's claims - it would be interesting to hit his SB3 with this program, I begin to wonder if the result may not actually be what you might expect... I recorded a 30s chunk, played out at 44k1. That means 1.3 million samples. 'This' (http://www.springerlink.com/content/q257543588327v45/) indicates the short-term stability of crystals is about 1 part in 10^9 but I don't know if that holds for stock parts. If it did I'd not expect to drift more than 1/1000 of a sampling period though I've had enough radio xtals that your could hear in a FM receiver if you ping the xtal with a screwdriver so they are vibration (music perhaps :) sensitive to some extent. It is true that th SB1 is the common element here and maybe I got a Friday afternoon box. I'll repeat on some other piece of digital audio gear to eliminate that last common variable. -- ermine ermine's Profile: http://forums.slimdevices.com/member.php?userid=12613 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I'd go further and say this program is of extremely limited practical usefulness, at least with external recordings (ie not made by itself). And I'll only give it that qualification rather than a full NFG because I haven't tried that. I recorded the output of the SqueezeBox 1 for 30s using Sound devices 702, then pressed red power off button and then play and play again to start track again. All the while recording the analogue out. Split resulting recording into first section and the second - no questions, no pack drill, these *are* the same thing. Stick this into audiodiffmaker. Result, perfectly audible if bassy track, a short-lived null at about 18s followed by perfectly audible track. Okay, perhaps Sound Devices clock is not stable enough. If that isn't, you average sound card isn't either... So I record simultaneously L of SB1 with the left of the same box but through the Arcam Black Box external DAC. One wired to the L channel of the SD702 and the other to the R channel. Now the clock must be synchronous either side, subject to the small latency spread (about 14 samples @ 44k1). Neither piece of kit is modded in any way. Resultant diff signal is about 30dB down with bass peaking at -15dB. Even I don't expect there to be so much difference between the two. I check for channels switched by repeating with the channels switched - this time diff signal sounds like classic stereo differencing and residual averages higher. OK so I probably got the channels right. For a final sanity check I take a phono y splitter and feed the same signal to both L and R of the recorder, via the selfsame cables. Now we're talking - residual down to -90, some peaking on LF to about -80. So at least satisfactory operation can be achieved if the recordings are not separated in time greatly So the Audio Critic guy is probably wrong in this case. Writing off the time-separated recordings as beyond the capability of the program, the simultaneous recording of the Squeezebox1 LH and the same feeding the Arcam BB1 LH show a difference. Audiodifference as recommended by Audio Critic to prove the converse indicates there is a significant difference at about -30dBFS. As a final sanity check I sent a 440Hz + 1kHz signal peaking at -3dBFS through both bits of kit to see if there was significant clipping or anything else obviously wrong, and both came through OK, though the spectrum of residual spurs were different. -- ermine ermine's Profile: http://forums.slimdevices.com/member.php?userid=12613 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I wouldn't draw any conclusions one way or the other from that. It sounds like your recording device simply isn't good enough, or there's some other problem somewhere. I tried this three times - with a SoundBlaster card in my PC, with a Tascam USB mic interface using the line-in, and with my Mac Book Pro's sound card. The recordings were of a stereo musical track, about 30s long, with one recording made with a FLAC source and one WAV (decoded on the SB - the idea was to debunk this nonsense about FLAC sounding worse than WAV). So the recordings were time-separated, and not even the same length (due to some variation in when I hit start and stop). In none of those cases did I get anything like what you're describing. The Tascam introduced a high-pitched whine that was audible in both the recording and the difference, so I didn't use it further. However both the soundcard recordings made difference files that were close to white noise. The mac recording had lower noise and a whiter spectrum, but both were at least 75 dB or so down. So my guess is your recording device has a problem of some kind. -- opaqueice opaqueice's Profile: http://forums.slimdevices.com/member.php?userid=4234 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
Okay, so I try this again, using my Toshiba laptop and another copy d/l just now. I have to rack the digital gain of the SB1 down to keep it within the input range of the line-in, but I do the same as before. I record the beginning of a track, stop, record it played again. The actual track used is different from the previous tests, I can't stand hearing the other one again. Difference signal peaks at about -35dBFS, this time mainly on the bass and cymbals. Difference is again perfectly recognizeable. Once again there is a short minimum at 18 seconds in, though the recognizeable song never actually goes away. This is done in the original mode the prog is intended, using itself to perform the recording. Then just for the hell of it I repeat with a HiMD recorder set to PCM mode and digitally transferred, and inserted in the same was as with the SD recorder. Residual peaks around -30DBFS mark. So now everything apart from my Squeezebox1 has been substituted - and each of these recordings should be exactly the same - no change between the successive recordings of gear, volume, moon phase, cables, incantations and curses... Now there are a number of possible conclusions: 1) I am lying, my results are the same as yours. Only I know that categorically, but I do have lots of recordings now :) 2) This program is duff - or I have downloaded two duff copies of this program v 1.10 from Audio Critic - one today and the other a couple of days ago 3) My SB1 system is playing up in the same way as Omega claims in this thread, ie its performance is not invariant in time. I'm damned if I can hear it, but then I haven't sat down to listen to a song repeated that hard 4) I have three independent duff recorders, which are mutually conspiring to give a similarly audible signal related problem. 5) Some other combination of effects of the highest audiophile voodoo But the one thing I cannot conclude from this is that this program shows me there is no variation in electronic pathways. It is just not repeatable enough - unless it is in fact repeatable and it is showing my SB1 has condition 3 and I simply don't have Omega's finely tuned hearing and/or my old audio system is masking the difference! I'm not quite ready for that leap, and I favour 2 over 3 at the moment. -- ermine ermine's Profile: http://forums.slimdevices.com/member.php?userid=12613 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
I've used it to compare WAV to FLAC playback through the analogue outs of my SB3. In that case the levels were identical, so that wasn't an issue. The difference file was white noise -85dB down or so, with the noise probably coming mostly from the recording (which was done using my MacBook's soundcard). There are several effects I can think of which could be creating a non-zero difference file in your case. The most obvious candidate is bad level matching - maybe audiodiffmaker doesn't do a great job with that. You can test that by changing the level of one of the files with another program (Audacity, say) and then run diffmaker on that modified file plus the other, and see if the difference file is the same as before (it should be if diffmaker works properly). You could also start with two identical copies of the same track, change the level of one, and then diff them - you should get something very close to zero. If that's not the cause another possibility is the recording is bad. One test of that would be to make two recordings of the same thing - like two recordings of the same track played over the SB1 - and diff them. The difference file will give you a sense of the resolution of the recording chain. If it's none of those it's probably really there. Incidentally that's not necessarily so surprising, even from the audio critic's point of view (which I more or less share). I had a DAC which distorted heavily when played at low volumes into a particular amp (due to an impedance mismatch, I think). At higher volume settings (on the amp that is) the distortion wasn't audible, but I suspect something as sensitive as audiodiffmaker would still have picked it up. So it's certainly not true that all digital sources sound the same always. The SB1 was pretty early in the product cycle - maybe it just has a bad analogue stage or something. Or maybe something in there distorts phase. Audiodiffmaker is an extremely sensitive test - much more sensitive than human hearing. For example nearly all audio systems mangle phase information totally (look at the step response of any multi-way speaker if you don't believe me). Phase is almost totally inaudible, so that doesn't matter much for sound, but audiodiffmaker will reveal phase distortions very very clearly (maybe that's happening in your case?). Polarity inversion, for example, would result in a difference file twice as loud as the original even though it's probably inaudible. So a positive result with diffmaker does NOT mean the files were audibly different, but a null result means they were not. -- opaqueice opaqueice's Profile: http://forums.slimdevices.com/member.php?userid=4234 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
opaqueice;219008 Wrote: The most obvious candidate is bad level matching - maybe audiodiffmaker doesn't do a great job with that. You can test that by changing the level of one of the files with another program (Audacity, say) and then run diffmaker on that modified file plus the other, and see if the difference file is the same as before (it should be if diffmaker works properly). Hmm. I take one of the recordings I made earlier, of the black box section I think, but it shouldn't matter. I then run this into Audition, lower the level by 10dB and compare it with the original. Audition is set to dither at 16bits. Diff needs amplifying by 60dB to reach 0dBFS and is a hissy copy of the original, with the bass jacked up massively. It is, however, perfectly easy to follow the song and hear the words, though there is also plently of signal-dependent noise. This is not what I expect to be the results of nulling out a digital 10dB attenuation, and starts to indicate to me that this program is not as useful as it is cracked up to be. There is, of course, the possibility that Adobe were incompetent in their implemenation of digital loss so jury's out. I then record the output of my SB1 again. Okay so I didn't match levels, I should have recorded two in succession I guess. I compare this with yesterdays. I expect this to give me a null. My section is 30s, a sort of null occurs at 17s and then the diff signal becomes bassy copy of the music. Almost as if the sampling rate changed slightly and one copy is phasing in and out. Okay so maybe Sound Devices should do better, but it's just not reasonable to expect a recorder to drift less than half a clock cycle over 30 seconds - that's less than one part in a million. Audio Critic's pal's software has to have a way to correct for that else the program isn't really useful in the real world. The impression I had gotten from the Audio Critic's cocky promotion of the prog was that this program would correct for linear distortions in time and in amplitude. I guess I can't complain too much, can't beat the $ price, but it has wasted some of my time... opaqueice;219008 Wrote: If it's none of those it's probably really there. Incidentally that's not necessarily so surprising, even from the audio critic's point of view (which I more or less share).). Yeah, no problem. We're all entitled to our religious viewpoints, I'm not even saying mine is right, it worked okay for me. Listening rather than reading specs bought me a stereo I've enjoyed for the last 15-20 years and the key bits even survived the change from records to CDs. What seems to be killing my enjoyment now is the rotten mastering of current pop/rock CDs :( I'd merely picked up on this program because it did seem to be really promising as a way of chasing down true nulls and/or proving the existence of subtle differences. But it seems to find too many differences to be useful to me. I'm not ready for sonic differences in mains leads etc and I was hoping to apply this to some of the more outlandish claims, or find there is something there even though that sort of thing isn't audible to me. Or maybe I am not licensed to drive diffmaker, maybe I need to believe in the invariance of electronic pathways before it'll play... At least I got a null with two copies of the same thing so I can't be doing too much wrong! Thanks for the pointers. I'll give it one more try recording the same thing twice immediately one after the other. I want to get at least one bona fide null out of this program before I toss it! -- ermine ermine's Profile: http://forums.slimdevices.com/member.php?userid=12613 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
Re: [SlimDevices: Audiophiles] How is Audiodiffmaker meant to work?
Hmm - so I guess the routine for level matching is just broken. In that case its utility is limited, although it should still be good for comparing FLAC to WAV, or digital cables, or two sources into the same DAC, or maybe interconnects. In any case (because of the phase issue plus other problems) all it can do reliably is put an upper bound on the difference between two signals. In your case that wasn't very useful. -- opaqueice opaqueice's Profile: http://forums.slimdevices.com/member.php?userid=4234 View this thread: http://forums.slimdevices.com/showthread.php?t=37352 ___ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles